AI Mock Interview Practice for Data Analysts & Scientists
Analyst and scientist loops test four distinct skills — SQL, statistical thinking, case framing, and stakeholder communication. Rehearse each separately with real questions, a worked case answer, and the rubric interviewers use.
The loop you're prepping for
Most data analysts & scientists loops share the same skeleton. Rehearse each round on its own — a single "general" mock trains you for none of them.
| Round | Length | What they score |
|---|---|---|
| Recruiter screen | 20–30 min | Tools you actually use daily, comp fit, one project story. |
| SQL / take-home | 45–60 min | Correctness, readability, edge cases (nulls, duplicates), efficient joins. |
| Product case | 45 min | Metric definition, hypothesis generation, prioritisation of drivers. |
| A/B test / stats | 45 min | Test design, power, common pitfalls (novelty, peeking, network effects). |
| Stakeholder communication | 30–45 min | Explaining a technical result to a non-technical audience with a clear ask. |
Real questions to practice — by round
SQL
- Find the second-highest salary per department.
- Compute 7-day rolling active users from an events table.
- Given orders + returns, compute net revenue by cohort month.
- Find users who did event A but never event B in the same session.
Product case / metrics
- Signups are up 20% but activation is flat. How do you diagnose?
- Define success for a new comments feature — 3 metrics.
- Weekly revenue dropped 4% — walk me through your investigation.
- How would you measure 'quality' of a search result?
A/B tests & statistics
- Design an A/B test for a new checkout button colour. What sample size do you need?
- The test shows +2% conversion with p=0.03 — do you ship?
- How do you handle network effects in a marketplace A/B test?
- Your treatment group has a 5% imbalance in a key covariate — what do you do?
Stakeholder / behavioural
- Tell me about a time you had to say 'the data doesn't support that'.
- Describe a dashboard nobody used — what would you do differently?
- Walk me through a project where your analysis changed a business decision.
Worked example
Question
Signups are up 20% week-over-week but activation is flat. How do you diagnose?
Strong sample answer
First I want to make sure the signal is real — I'd verify no tracking change, no bot influx, no marketing test skewing the top of funnel. Assume that's clean. Now the framing: activation is defined as X within Y days of signup. Two ways activation can stay flat while signups rise — (a) the new signups are lower-quality than baseline (channel mix shift), or (b) something in the activation step regressed at the same time. Cut 1 — channel: split activation rate by acquisition source for the last 4 weeks. If paid social jumped from 20% to 45% of signups and paid social activates at 8% vs. organic at 30%, that mathematically explains a flat overall rate. Cut 2 — cohort: plot activation rate by signup week, holding channel constant. If organic activation is still 30%, it's a mix problem, not a product problem. If organic dropped from 30% to 22%, something regressed — look at deploy log, onboarding funnel step drop-off, and platform mix. Cut 3 — funnel: for the new signups that didn't activate, which step did they fall off? If it's step 1, we may be showing an overloaded onboarding to a colder audience. If it's the final "confirm" step, likely a bug or a form change. I'd bring back a one-page write-up: the root cause with the number that proves it, one recommended action, and one metric to watch for reversal. The mistake to avoid — a 12-tab notebook. Nobody reads that.
The rubric interviewers use
Hypothesis generation
You named 2–3 plausible causes before running any query. Not shotgun-analysing.
Metric literacy
You clarified the metric definition. You know the difference between rate and count, cohort and snapshot.
Rigor without over-engineering
You picked the smallest analysis that could answer the question. Time-boxed.
Communication
You'd deliver one clear recommendation, not a data dump. Stakeholders remember conclusions, not code.
Tips that actually move your score
- For SQL rounds, narrate what you're doing before typing — interviewers score the plan more than the syntax.
- In case rounds, write the metric definition on paper before you start. Ambiguous metric = ambiguous case.
- For A/B test questions, always name the guardrail metric — 'we ship if primary is +X and no guardrail regressed'. That single sentence lifts scores.
- In stakeholder rounds, lead with the ask ('I want to sunset feature X') and back-fill with data. Non-technical audiences hire clarity.
Frequently asked questions
How much SQL do I need for a data analyst interview?
Comfort with joins, window functions, CTEs, and null-handling edge cases. If you can compute retention curves and funnel drop-off in SQL under 20 minutes, you're ready.
Do data scientist interviews still ask A/B test questions?
Yes — even for ML-heavy roles. Every shipped model becomes an experiment; interviewers screen out candidates who conflate significance with impact.
How do I practice case interviews as an analyst alone?
Take a Kaggle dataset, invent a stakeholder question, and write a 1-page memo. Do 5 of these and case rounds stop feeling like guessing.
Also read: STAR method interview questions & examples · Mock interview practice hub.
Practice these questions with Upla
Role-calibrated AI mock interviews with instant scoring on structure, specificity, impact, and delivery.
Start a mock interview